We speed up the rendering of participating media with Virtual Ray Lights (VRLs) by clustering them in a preprocessing step. A subset of representative VRLs is then sampled from the clustering, which is used for the final rendering. By performing a full variance analysis, we can explicitly estimate the convergence rate of the rendering process and automatically find the locally ideal number of clusters to maximize efficiency. Overall, we report speed-up factors ranging from 13 to 16 compared to unclustered rendering.
Authors: Roald Frederickx, Pieterjan Bartels, Philip Dutré